Confidence In Predictions Research Articles

Estimation of porosity from well logs and seismic using artificial neural network

The Artificial Neural Network (ANN) is widely used to map and estimate reservoir properties. Since ANN has the ability of non-linear computing and self-error correction, it serves as an alternative method to enhance reservoir characterization by improving reservoir properties' prediction. This study employed an integrated approach where seismic, and well log data are used alongside ANN to evaluate petrophysical properties in Field X, South Caspian Basin. The study field is situated in the South Caspian Basin, which developed during Tertiary-Quaternary. The South Caspian Basin covers the southern part of the Caspian Basin, coastal regions of eastern Azerbaijan, northern Iran, and western Turkmenistan. The field is geologically situated on an elongated and multi-crest, anticlinal feature, better known as Apsheron Sill. Available data set consists of log data from four wells: X1, X2, X3, X4, and 3D pre-stack seismic data. Firstly, the reservoir properties were calculated from the well log data. The simultaneous inversion was carried out to obtain elastic rock properties like the P-impedance, S-impedance, density. After that, elastic properties were used to obtain 3D dimensional SQp and SQs attributes. Furthermore, the Radial basis function Neural Network (RBFN) model was created with optional inputs: volume of P-impedance, S-impedance, density, SQp, SQs and property logs and outputs: porosity. Moreover, this research work demonstrates an application of SQp and SQs attributes for properties prediction using neural network. The neural network's primary purpose is to determine a relationship between obtained volumes and reservoir parameters, porosity at well locations. The RBFN model is successfully trained and validated on the field data. The results demonstrated an excellent correlation between actual property logs. They predicted properties, which gives confidence for spatial prediction in areas where well logs are not available.

New England Shelf Break Acoustic (NESBA) experiment: Stochastic acoustic prediction

Sonar performance prediction has been a topic of interest for decades. Predictions are based on acoustic propagation modeling that is typically deterministic, i.e., lacking quantified confidence or uncertainty bounds. While this may be adequate in some locations, in oceanographically complex regions it can result in misleading predictions and conclusions that sonar performance predictions are unreliable. A better approach includes explicit calculations of the underlying source, receiver, and environmental uncertainties to provide a prediction confidence level. In oceanographically complex areas, the resulting uncertainties are likely to be high; however, knowledge of this allows uncertainties to be managed. The NESBA Signals and Noise experiment was conducted in April–May 2021. The goal of the experiment was to assess the potential for sonar prediction effectiveness gains given improved environmental awareness. This talk addresses the NESBA sub-goal of demonstrating that high-resolution ocean modeling with ensemble forecasts (in this case, Regional NCOM) could be leveraged for stochastic acoustic predictions, resulting in increased accuracy in performance predictions with quantified uncertainty bounds. NESBA results will be presented leveraging the high-resolution ocean model forecasts with ensembles run concurrently with NESBA and the numerous temperature/salinity profiles taken during the experiment. [Work supported by the Office of Naval Research.]

DIII-D research advancing the physics basis for optimizing the tokamak approach to fusion energy

DIII-D physics research addresses critical challenges for the operation of ITER and the next generation of fusion energy devices. This is done through a focus on innovations to provide solutions for high performance long pulse operation, coupled with fundamental plasma physics understanding and model validation, to drive scenario development by integrating high performance core and boundary plasmas. Substantial increases in off-axis current drive efficiency from an innovative top launch system for EC power, and in pressure broadening for Alfven eigenmode control from a co-/counter-I p steerable off-axis neutral beam, all improve the prospects for optimization of future long pulse/steady state high performance tokamak operation. Fundamental studies into the modes that drive the evolution of the pedestal pressure profile and electron vs ion heat flux validate predictive models of pedestal recovery after ELMs. Understanding the physics mechanisms of ELM control and density pumpout by 3D magnetic perturbation fields leads to confident predictions for ITER and future devices. Validated modeling of high-Z shattered pellet injection for disruption mitigation, runaway electron dissipation, and techniques for disruption prediction and avoidance including machine learning, give confidence in handling disruptivity for future devices. For the non-nuclear phase of ITER, two actuators are identified to lower the L–H threshold power in hydrogen plasmas. With this physics understanding and suite of capabilities, a high poloidal beta optimized-core scenario with an internal transport barrier that projects nearly to Q = 10 in ITER at ∼8 MA was coupled to a detached divertor, and a near super H-mode optimized-pedestal scenario with co-I p beam injection was coupled to a radiative divertor. The hybrid core scenario was achieved directly, without the need for anomalous current diffusion, using off-axis current drive actuators. Also, a controller to assess proximity to stability limits and regulate β N in the ITER baseline scenario, based on plasma response to probing 3D fields, was demonstrated. Finally, innovative tokamak operation using a negative triangularity shape showed many attractive features for future pilot plant operation.

Rectangular Confidence Regions and Prediction Regions in Multivariate Calibration

The multivariate calibration problem deals with inference concerning an unknown value of a covariate vector based on an observation on a response vector. Two distinct scenarios are considered in the multivariate calibration problem: controlled calibration where the covariates are non-stochastic, and random calibration where the covariates are random. Under controlled calibration, a problem of interest is the computation of a confidence region for the unknown covariate vector. Under random calibration, the problem is that of computing a prediction region for the covariate vector. Assuming the standard multivariate normal linear regression model, rectangular confidence and prediction regions are derived using a parametric bootstrap approach. Numerical results show that the regions accurately maintain the coverage probabilities. The results are illustrated using examples. The regions currently available in the literature are all ellipsoidal, and this work is the first attempt to derive rectangular regions.

ANN-Based Inverse Goal-Oriented Design Method for Targeted Final Properties of Materials

Designing materials for targeted materials properties is the key to tackle the demands for personalized consumer products. The deficiency in the existing linear and nonlinear correlation methods attributed to simplifying assumptions and idealizations, nondeterministic simulations, and limited experimental data due to heavy computational time and cost, necessitates a design method that provides sufficient confidence to designers in decision making. To address this requirement, we propose, in this paper, an inverse goal-oriented materials design method supported by the design space exploration framework (DSEF). Keeping in view the accuracy and precision in the prediction confidence of machine learning-based methods, we developed an Artificial Neural Network based prediction model that supports DSEF. The proposed method for materials design can help designers to (1) explore PSPP spaces starting from end property requirements, (2) adjust the errors being propagated in the PSPP chain as well as in the predictions made by the model, and (3) timely adjust model parameters of the prediction model for accurate predictions. The efficacy of the method is illustrated for the hot stamping process to produce structural components from ultrahigh-strength steels (UHSS). The proposed method and prediction model are generic and applicable to any sequential manufacturing process to realize an end product.

Deep Learning-Based Monocular 3D Object Detection with Refinement of Depth Information.

Recently, the research on monocular 3D target detection based on pseudo-LiDAR data has made some progress. In contrast to LiDAR-based algorithms, the robustness of pseudo-LiDAR methods is still inferior. After conducting in-depth experiments, we realized that the main limitations are due to the inaccuracy of the target position and the uncertainty in the depth distribution of the foreground target. These two problems arise from the inaccurate depth estimation. To deal with the aforementioned problems, we propose two innovative solutions. The first is a novel method based on joint image segmentation and geometric constraints, used to predict the target depth and provide the depth prediction confidence measure. The predicted target depth is fused with the overall depth of the scene and results in the optimal target position. For the second, we utilize the target scale, normalized with the Gaussian function, as a priori information. The uncertainty of depth distribution, which can be visualized as long-tail noise, is reduced. With the refined depth information, we convert the optimized depth map into the point cloud representation, called a pseudo-LiDAR point cloud. Finally, we input the pseudo-LiDAR point cloud to the LiDAR-based algorithm to detect the 3D target. We conducted extensive experiments on the challenging KITTI dataset. The results demonstrate that our proposed framework outperforms various state-of-the-art methods by more than 12.37% and 5.34% on the easy and hard settings of the KITTI validation subset, respectively. On the KITTI test set, our framework also outperformed state-of-the-art methods by 5.1% and 1.76% on the easy and hard settings, respectively.

A perspective on calibration and application of DEM models for simulation of industrial bulk powder processes

In recent years, there has been increased focus on the calibration of discrete element method (DEM) models to gain improved confidence in the simulation predictions. However, most recent reports have focused on calibration for real particles of idealized shape (e.g., spherical) or large size (e.g., >1 mm), presented complex calibration procedures that may be difficult to adopt in an industrial setting, lack documented experimental validation of the approach, or apply to only a specific material of interest. In this article, we present an industrial perspective on application of DEM models centered on a generalized calibration approach for fine powders with low-to-moderate cohesion that is simple, rapid, and requires minimal experimentation. The approach, based on previous work, is formalized here and utilizes non-rotating spherical parcels to represent a small volume of powder, and in many cases, the simulation results are reasonably predictive of bulk powder flow behavior in dense systems. The assumptions and limitations of the approach, along with some guidelines for application are discussed. Finally, several examples from the literature demonstrating success of the approach in industrially-relevant powder processes with comparison to either commonly accepted theory or experimental data are presented.

IDC: quantitative evaluation benchmark of interpretation methods for deep text classification models

Recent advances in deep neural networks have achieved outstanding success in natural language processing tasks. Interpretation methods that provide insight into the decision-making process of these models have received an influx of research attention because of the success and the black-box nature of the deep text classification models. Evaluation of these methods has been based on changes in classification accuracy or prediction confidence when removing important words identified by these methods. There are no measurements of the actual difference between the predicted important words and humans’ interpretation of ground truth because of the lack of interpretation ground truth. A large publicly available interpretation ground truth has the potential to advance the development of interpretation methods. Manual labeling important words for each document to create a large interpretation ground truth is very time-consuming. This paper presents (1) IDC, a new benchmark for quantitative evaluation of interpretation methods for deep text classification models, and (2) evaluation of six interpretation methods using the benchmark. The IDC benchmark consists of: (1) Three methods that generate three pseudo-interpretation ground truth datasets. (2) Three performance metrics: interpretation recall, interpretation precision, and Cohen’s kappa inter-agreement. Findings: IDC-generated interpretation ground truth agrees with human annotators on sampled movie reviews. IDC identifies Layer-wise Relevance Propagation and the gradient-by-input methods as the winning interpretation methods in this study.

PBK modelling of topical application and characterisation of the uncertainty of Cmax estimate: A case study approach

Combined with in vitro bioactivity data, physiologically based kinetic (PBK) models has increasing applications in next generation risk assessment for animal-free safety decision making. A tiered framework of building PBK models for such application has been developed with increasing complexity and refinements, as model parameters determined in silico, in vitro, and with human pharmacokinetic data become progressively available. PBK modelling has been widely applied for oral/intravenous administration, but less so on topically applied chemicals. Therefore, building PBK models for topical applications and characterizing their uncertainties in the tiered approach is critical to safety decision making. The purpose of this study was to assess the confidence of PBK modelling of topically applied chemicals following the tiered framework, using non-animal methods derived parameters. Prediction of maximum plasma concentration (Cmax) and area under the curve were compared to observed kinetics from published dermal clinical studies for five chemicals (diclofenac, salicylic acid, coumarin, nicotine, caffeine). A bespoke Bayesian statistical model was developed to describe the distributions of Cmax errors between the predicted and observed data. We showed a general trend that confidence in model predictions increases when more quality in vitro data, particularly those on hepatic clearance and dermal absorption, are available as model input. The overall fold error distributions are useful for characterizing model uncertainty. We concluded that by identifying and quantifying the uncertainties in the tiered approach, we can increase the confidence in using PBK modelling to help make safety decisions on topically applied chemicals in the absence of human pharmacokinetic data.

Unsupervised Machine Learning Techniques for Detecting PLC Process Control Anomalies

The security of programmable logic controllers (PLCs) that control industrial systems is becoming increasingly critical due to the ubiquity of the Internet of Things technologies and increasingly nefarious cyber-attack activity. Conventional techniques for safeguarding PLCs are difficult due to their unique architectures. This work proposes a one-class support vector machine, one-class neural network interconnected in a feed-forward manner, and isolation forest approaches for verifying PLC process integrity by monitoring PLC memory addresses. A comprehensive experiment is conducted using an open-source PLC subjected to multiple attack scenarios. A new histogram-based approach is introduced to visualize anomaly detection algorithm performance and prediction confidence. Comparative performance analyses of the proposed algorithms using decision scores and prediction confidence are presented. Results show that isolation forest outperforms one-class neural network, one-class support vector machine, and previous work, in terms of accuracy, precision, recall, and F1-score on seven attack scenarios considered. Statistical hypotheses tests involving analysis of variance and Tukey’s range test were used to validate the presented results.

Near-Field Anisotropic Mesh Adaptation for the Third AIAA Sonic Boom Workshop

Sonic boom prediction is critical for the development of low-boom supersonic aircraft and encouraging a replacement of the overland commercial supersonic ban with a certification standard. The Third AIAA Sonic Boom Workshop near-field test cases provide a unique opportunity to review anisotropic adaptive-mesh strategies for complex geometries with important propulsion interaction effects. Anisotropic metrics formulated to control estimated Mach interpolation and integrated near-field pressure signature (goal-oriented) errors are compared with the results obtained on workshop-provided meshes. Detailed flow solutions and mesh renderings are combined with computational schlieren images to reveal the complex interactions present in a shock–plume interaction wind tunnel model and a prototype of the Low-Boom Flight Demonstrator. A favorable comparison of independent implementations of the entire simulation process provides evidence that each process is implemented correctly and sets a solid foundation for confidence in the near-field prediction of each method.

A comprehensive toolbox for the gamma distribution: The gammadist package

The gamma distribution is one of the most important parametric models in probability theory and statistics. Although a multitude of studies have theoretically investigated the properties of the gamma distribution in the literature, there is still a serious lack of tailored statistical tools to facilitate its practical applications. To fill the gap, this paper develops a comprehensive R package for the gamma distribution. In specific, the R package focuses on the following three important tasks: generate the gamma random variables, estimate the model parameters, and construct statistical limits, including confidence limits, prediction limits, and tolerance limits based on the gamma random variables. The proposed package encompasses the state-of-the-art methods of the gamma distribution in the literature and its usage is illustrated by a real application.

Future wildfire extent and frequency determined by the longest fire-conducive weather spell

Great efforts have been made to understand the impacts of a changing climate on fire activity; however, a reliable approach with high prediction confidence has yet to be found. By establishing linkages between the longest duration of fire-conducive weather spell and fire activity parameters, this study projected annual area burned (AAB), annual number of fires (ANF), and annual maximum fire size (MFS) into the future. We found that even though the rates of change differ, the spatial pattern of changes for all three parameters are similar by Canadian ecozone. Areas with the lowest fire activity may see higher rates of change in comparison to high fire activity areas. By end of the century, the changes of AAB and MFS for the study area are projected to be about four and five times that of the baseline respectively, while ANF could almost double.

Development of CNN models for the enteral feeding tube positioning assessment on a small scale data set

BackgroundEnteral nutrition through feeding tubes serves as the primary method of nutritional supplementation for patients unable to feed themselves. Plain radiographs are routinely used to confirm the position of the Nasoenteric feeding tubes the following insertion and before the commencement of tube feeds. Convolutional neural networks (CNNs) have shown encouraging results in assisting the tube positioning assessment. However, robust CNNs are often trained using large amounts of manually annotated data, which challenges applying CNNs on enteral feeding tube positioning assessment.MethodWe build a CNN model for feeding tube positioning assessment by pre-training the model under a weakly supervised fashion on large quantities of radiographs. Since most of the model was pre-trained, a small amount of labeled data is needed when fine-tuning the model for tube positioning assessment. We demonstrate the proposed method using a small dataset with 175 radiographs.ResultThe experimental result shows that the proposed model improves the area under the receiver operating characteristic curve (AUC) by up to 35.71% , from 0.56 to 0.76, and 14.49% on the accuracy, from 0.69 to 0.79 when compared with the no pre-trained method. The proposed method also has up to 40% less error when estimating its prediction confidence.ConclusionOur evaluation results show that the proposed model has a high prediction accuracy and a more accurate estimated prediction confidence when compared to the no pre-trained model and other baseline models. The proposed method can be potentially used for assessing the enteral tube positioning. It also provides a strong baseline for future studies.

Rheological based interpretation of shotcrete pumpability and shootability

In this article, particle size distribution in shotcrete and the diverse effects of admixtures on gradation were discussed. Although admixtures can enhance concrete rheological behavior, they introduce uncertainty into mixture behavior and may lead to segregation, pumping line blockage, and low shootability. Moreover, the assessment of these factors using the Chapdelaine and Burns concepts showed divergence from practical pumpability. We introduced a method to analyze shotcrete behavior using a rheological method that was found more effective. As a result, a model for predicting the build-up thickness was proposed using a statistical method. In addition, a zonal effectiveness plot that could accurately characterize pumpability and shootability, overcoming the uncertainties associated with normal shotcrete analysis methods, was developed using confidence ellipses and prediction intervals that can be used in practical applications in the wet-mixed shotcrete field.

Pedestal analysis of MAST ELMy regimes

Mega Ampere Spherical Tokamak (MAST) pedestal data has been analysed, where a pedestal database of 892 shots was obtained, using the upgraded MAST Thomson scattering (TS) diagnostic. Various edge localised mode (ELM) types are discussed, where characteristics and trends of MAST pedestals are shown. The data from the upgraded TS diagnostic confirms pedestal characteristics found in earlier analysis, using previous TS systems. Using the database, calculations of the bootstrap current are obtained using the different analytical formulae (Sauter and Redl), and comparisons are performed. The upgraded MAST TS system now spans the full plasma mid-plane, such that direct comparisons between inboard and outboard pedestals can be obtained, and asymmetries in the density pedestal width were found. To increase confidence in spherical tokamak pedestal predictions, the assumptions of Europed have been validated from the MAST pedestal data, and a value for the kinetic ballooning mode constraint () has been obtained. The first spherical tokamak pedestal predictions were performed in Europed and compared to experimental values. Using C = 0.145 the temperature pedestal height was predicted to within 10% of the experimental value. In addition type II ELMs on MAST are analysed, and stability analysis and parameter scans have been performed using ESSIVE. Similar magneto-hydrodynamic (MHD) stability properties are observed for type I and type II ELMs, originating from the mixed ELM regimes; it is therefore difficult to distinguish these ELMs using the ideal MHD codes.

In Vitro - in Vivo Extrapolation of Hepatic Clearance in Preclinical Species.

Accurate prediction of human clearance is of critical importance in drug discovery. In this study, in vitro - in vivo extrapolation (IVIVE) of hepatic clearance was established using large sets of compounds for four preclinical species (mouse, rat, dog, and non-human primate) to enable better understanding of clearance mechanisms and human translation. In vitro intrinsic clearances were obtained using pooled liver microsomes (LMs) or hepatocytes (HEPs) and scaled to hepatic clearance using the parallel-tube and well-stirred models. Subsequently, IVIVE scaling factors (SFs) were derived to best predict in vivo clearance. The SFs for extended clearance classification system (ECCS) class 2/4 compounds, involving metabolic clearance, were generally small (≤ 2.6) using both LMs and HEPs with parallel-tube model, with the exception of the rodents (~ 2.4-4.6), suggesting in vitro reagents represent in vivo reasonably well. SFs for ECCS class 1A and 1B are generally higher than class 2/4 across the species, likely due to the contribution of transporter-mediated clearance that is under-represented with in vitro reagents. The parallel-tube model offered lower variability in clearance predictions over the well-stirred model. For compounds that likely demonstrate passive permeability-limited clearance in vitro, rat LM predicted in vivo clearance more accurately than HEP. This comprehensive analysis demonstrated reliable IVIVE can be achieved using LMs and HEPs. Evaluation of clearance IVIVE in preclinical species helps to better understand clearance mechanisms, establish more reliable IVIVE in human, and enhance our confidence in human clearance and PK prediction, while considering species differences in drug metabolizing enzymes and transporters.

Epsilon Consistent Mixup: Structural regularization with an adaptive consistency‐interpolation trade‐off

In this paper, we propose Epsilon Consistent Mixup (∈mu). ∈mu is a data‐based structural regularization technique that combines Mixup's linear interpolation with consistency regularization in the Mixup direction, by compelling a simple adaptive tradeoff between the two. This learnable combination of consistency and interpolation induces a more flexible structure on the evolution of the response across the feature space and is shown to improve semi‐supervised classification accuracy on the SVHN and CIFAR10 benchmark datasets, yielding the largest gains in the most challenging low label‐availability scenarios. Empirical studies comparing ∈mu and Mixup are presented and provide insight into the mechanisms behind ∈mu's effectiveness. In particular, ∈mu is found to produce more accurate synthetic labels and more confident predictions than Mixup.

From Atmospheric Waves to Heatwaves: A Waveguide Perspective for Understanding and Predicting Concurrent, Persistent, and Extreme Extratropical Weather

Abstract A notable number of high-impact weather extremes have occurred in recent years, often associated with persistent, strongly meandering atmospheric circulation patterns known as Rossby waves. Because of the high societal and ecosystem impacts, it is of great interest to be able to accurately project how such extreme events will change with climate change, and to predict these events on seasonal-to-subseasonal (S2S) time scales. There are multiple physical links connecting upper-atmosphere circulation patterns to surface weather extremes, and it is asking a lot of our dynamical models to accurately simulate all of these. Subsequently, our confidence in future projections and S2S forecasts of extreme events connected to Rossby waves remains relatively low. We also lack full fundamental theories for the growth and propagation of Rossby waves on the spatial and temporal scales relevant to extreme events, particularly under strongly nonlinear conditions. By focusing on one of the first links in the chain from upper-atmospheric conditions to surface extremes—the Rossby waveguide—it may be possible to circumvent some model biases in later links. To further our understanding of the nature of waveguides, links to persistent surface weather events and their representation in models, we recommend exploring these links in model hierarchies of increasing complexity, developing fundamental theory, exploiting novel large ensemble datasets, harnessing deep learning, and increased community collaboration. This would help increase understanding and confidence in both S2S predictions of extremes and of projections of the impact of climate change on extreme weather events.

Multiple imputation of multibeam angular response data for high resolution full coverage seabed mapping

Acoustic data collected by multibeam echosounders (MBES) are increasingly used for high resolution seabed mapping. The relationships between substrate properties and the acoustic response of the seafloor depends on the acoustic angle of incidence and the operating frequency of the sonar, and these dependencies can be analysed for discrimination of benthic substrates or habitats. An outstanding challenge for angular MBES mapping at a high spatial resolution is discontinuity; acoustic data are seldom represented at a full range of incidence angles across an entire survey area, hindering continuous spatial mapping. Given quantifiable relationships between MBES data at various incidence angles and frequencies, we propose to use multiple imputation to achieve complete estimates of angular MBES data over full survey extents at a high spatial resolution for seabed mapping. The primary goals of this study are (i) to evaluate the effectiveness of multiple imputation for producing accurate estimates of angular backscatter intensity and substrate penetration information, and (ii) to evaluate the usefulness of imputed angular data for benthic habitat and substrate mapping at a high spatial resolution. Using a multi-frequency case study, acoustic soundings were first aggregated to homogenous seabed units at a high spatial resolution via image segmentation. The effectiveness and limitations of imputation were explored in this context by simulating various amounts of missing angular data, and results suggested that a substantial proportion of missing measurements (> 40%) could be imputed with little error using Multiple Imputation by Chained Equations (MICE). The usefulness of imputed angular data for seabed mapping was then evaluated empirically by using MICE to generate multiple stochastic versions of a dataset with missing angular measurements. The complete, imputed datasets were used to model the distribution of substrate properties observed from ground-truth samples using Random Forest and neural networks. Model results were pooled for continuous spatial prediction and estimates of confidence were derived to reflect uncertainty resulting from multiple imputations. In addition to enabling continuous spatial prediction, the high-resolution imputed angular models performed favourably compared to broader segmentations or non-angular data.